Joel Dubin, Ph.D.

Assistant Professor of Biostatistics

Department of Epidemiology and Public Health

School of Medicine

Yale University

 

 

 

Correlation for multivariate longitudinal data

 

 

In this presentation, we will describe methods to handle multivariate longitudinal data or multivariate responses that are followed repeatedly over time. These data are viewed as realizations of a random process.  Dependency between the various components of the response is characterized by a non-parametric correlation technique which we refer to as dynamical correlation.  The methods utilize smoothed curves constructed from the original data.  The assessment of the dynamics of the underlying processes also includes the consideration of derivatives and of time lags. Our methods are illustrated with data on five acute phase blood proteins measured longitudinally for a sample of patients requiring hemodialysis. Simulation results will also be presented.